MEMRISTORS 2025

Optimizing HfO₂ Memristors with an Al₂O₃ Barrier: Toward Reliable Non-Volatile Memory

  • Shooshtari, Mostafa (Instituto de Microelectrónica de Sevilla, IMS)
  • CAMPABADAL, FRANCESCA (Instituto de Microelectrònica de Barcelona, I)
  • LINARES-BARRANCO, BERNABE (Instituto de Microelectrónica de Sevilla, IMS)

Please login to view abstract download link

Hafnium oxide (HfO₂)-based memristors are promising candidates for next-generation non-volatile memory and neuromorphic computing applications. However, their performance is often limited by variability in switching characteristics, leakage currents, and retention issues. In this study, we investigate the impact of introducing a thin Al₂O₃ interfacial layer into HfO₂-based memristors to improve their performance. Two device structures, W/HfO₂/Ti/TiN and W/HfO₂/Al₂O₃/Ti/TiN, were fabricated and compared in terms of resistive switching behavior, endurance, and retention. The addition of the Al₂O₃ layer significantly enhances the device’s stability and uniformity by reducing leakage currents and improving oxygen vacancy confinement. The W/HfO₂/Al₂O₃/Ti/TiN device demonstrated a higher forming voltage (~4.9V vs. 3.2V) and a more stable resistive switching process compared to the single-layer HfO₂ device. Additionally, the Al₂O₃ layer increases the ON/OFF ratio, reduces cycle-to-cycle variability, and enhances the endurance of the device over 100 switching cycles. These improvements are attributed to the higher bandgap and insulating properties of Al₂O₃, which effectively modulate the electric field and suppress uncontrolled vacancy migration. The figure presents the I-V characteristics of both devices, highlighting the improved switching stability of the Al₂O₃-based structure. The W/HfO₂/Al₂O₃/Ti/TiN device exhibits a more defined resistive switching window, with lower leakage currents in the high-resistance state (HRS). These results demonstrate that incorporating an Al₂O₃ interfacial layer is a viable approach to overcoming the limitations of HfO₂-based memristors, making them more suitable for non-volatile memory and neuromorphic computing.